Fine-tuning vs. prompting
Fine-tuning means actually retraining a model further on your own examples, permanently changing its weights. Prompting means just giving good instructions/context at inference time without touching the model at all. For almost all product needs today, good prompting (plus RAG) beats fine-tuning - it's cheaper, faster to iterate, and doesn't require ML infrastructure.
Fine-tuning makes sense in narrow cases: you need a very specific consistent style/format at massive scale, you have thousands of high-quality labeled examples, and prompting genuinely can't achieve the consistency you need. It's expensive, slower to iterate (need to retrain to change behavior), and not reversible the way editing a prompt is. Most companies people assume are "fine-tuning" are actually just prompting well with RAG.
Advisory exercise: next time you hear a startup claim "we fine-tuned our own model," ask (mentally or literally) whether prompting + RAG could have achieved the same result cheaper - this is a genuinely useful skeptical instinct for evaluating AI product claims.
For most real product needs today, which usually wins?